A hybrid molecular-imaging model for high-accuracy early colorectal cancer diagnosis.

Background: Early and accurate detection of colorectal cancer (CRC) is crucial for improving patient survival; however, current screening methods often suffer from high false-negative rates, hindering timely diagnosis and treatment. This study aims to develop an innovative dual-path strategy that integrates molecular biomarkers with artificial intelligence (AI)-driven imaging techniques to enhance CRC detection accuracy and overcome limitations in existing screening methods. Methods: We utilized transcriptomic data from the Gene Expression Omnibus (GEO) database to identify nine key molecular biomarkers associated with CRC, including CDC25B and TEAD4, through differential expression analysis. Machine learning algorithms were employed to assess the diagnostic performance of these biomarkers. In parallel, an edge-aware Mamba-enhanced transformer network (EMT-Net) was developed for imaging segmentation, tested on the Computer Vision Center-Clinic Database (CVC-ClinicDB). Results: The molecular biomarkers showed significant diagnostic potential, achieving an area under the receiver operating characteristic curve (AUROC) of 0.987 in independent validation. The EMT-Net model demonstrated superior segmentation performance compared to current state-of-the-art methods on the CVC-ClinicDB, showing improved accuracy and precision in CRC detection. Conclusions: By combining molecular biomarker analysis with advanced imaging segmentation, our dual-path strategy offers complementary advantages: biological insights from molecular data and clinical precision from imaging techniques. This integrated approach shows exceptional cross-dataset robustness, with significant potential to enhance early CRC detection in clinical practice.